AI SDLC
What AI SDLC means, how the kit implements it, and the three phases of the AI Development Lifecycle.
What is AI SDLC?
AI SDLC (AI Software Development Lifecycle) is the application of AI agents across the full software development lifecycle — from inception to operations — with structured human oversight at key decision points.
It is distinct from simply "using AI to write code." In AI SDLC, AI agents participate in every phase: discovery, specification, implementation, testing, review, and operations. The human role shifts from doing the work to validating the work at critical gates.
The three phases of AI-DLC
The AI Development Lifecycle is organized in three macro-phases:
Phase 1 — Inception
Goal: transform a raw idea or a defined problem into a structured, validated specification.
| Step | Who | Output |
|---|---|---|
| Idea refinement | 🧭 Discovery | idea.md |
| Functional spec | 🗂️ PM | non-technical-spec.md |
| Product requirements | 🧑💼 Tech Lead | PRD.md |
| Technical spec | 🏗️ Architect | technical-spec.md |
| Epic breakdown | 🏗️ Architect | epics.md |
The Inception phase ends when a human approves epics.md — a sequenced, independently deliverable breakdown of the entire product.
Phase 2 — Construction
Goal: implement each epic with full spec traceability and human review gates.
For each epic:
| Step | Who | Output |
|---|---|---|
| Epic spec preparation | 🏗️ Architect | spec-epic-N.md, PRD.md, epic-N.md |
| Implementation | 🛠️ Implementer | Code + decisions-log.md |
| Testing | 🧪 QA | Test report |
| Review | 🔎 Reviewer | Review report |
| Human approval | Human | Merge decision |
The Construction phase repeats for every epic.
Phase 3 — Operations
Goal: close each epic safely into production and maintain project-wide context.
| Step | Who | Output |
|---|---|---|
| Deploy preparation | 🚀 Ops | ops-epic-N.md |
| Production validation | Human | Merge + deploy |
| Context sync | 🏗️ Architect | CONTEXT.md (updated) |
| Incident triage | 🚀 Ops | incident-log.md |
How the ai-sdlc-kit implements these concepts
| AI-DLC concept | ai-sdlc-kit implementation |
|---|---|
| Inception phase | Flow A — Discovery phase + /discovery-* prompts |
| Construction phase | Spec phase + /epic-init + /task-implement + /task-review |
| Operations phase | /epic-close + /context-sync + /ops-triage |
| Human oversight | HITL checkpoints after every generated artefact |
| Context memory | CONTEXT.md — read by all agents, updated after each epic |
| Audit trail | decisions-log.md — append-only ADR record per epic |
The principle: AI plans, human validates
The AI SDLC Kit is built on a single organizational principle:
AI agents are excellent at generating structured content from structured input. Humans are essential for validating that the content is correct in context.
AI agents in the kit do not carry:
- Product strategy and business context
- Risk tolerance and organizational constraints
- Team capacity and delivery reality
- The lived experience of what "done" means for this specific project
The HITL checkpoints exist precisely to inject that missing context at the moments it matters most — before each artefact becomes the input to the next stage.
AI SDLC vs. traditional SDLC
| Dimension | Traditional SDLC | AI SDLC |
|---|---|---|
| Specification | Written by humans | Co-authored by AI, validated by humans |
| Implementation | Written by humans | Generated by AI under spec constraints |
| Review | Manual peer review | AI-assisted adherence check + human final call |
| Documentation | Often deferred | Generated continuously as a byproduct |
| Context persistence | Tribal knowledge | CONTEXT.md — structured, queryable, agent-readable |
| Speed | Bounded by human bandwidth | Faster generation, same validation discipline |
The gain is speed and structure. The human responsibility does not shrink — it focuses on the decisions that matter.